Ecommerce AI Outbound Calls: Orders, Abandoned Checkouts and Delivery

When a call belongs in ecommerce

Ecommerce is a messaging business. Customers expect email and notifications, they do not expect the phone, and a call from an online shop is unusual enough to be either impressive or intrusive depending entirely on why it happened.

That makes the selection criterion unusually strict. A call is justified when there is a decision to be made, a problem to solve, or money at risk — and specifically when a message has failed or would be too slow. It is not justified for anything a notification handles perfectly well, which is most things.

Three use cases meet that bar for most online retailers: confirming orders where the risk of fraud or error is real, recovering abandoned checkouts of sufficient value, and handling delivery problems. This article covers what each call should do and, importantly, when each should not happen at all.

Use case 1: order confirmation

Confirming orders by phone is standard practice in some markets and unnecessary in others. The determining factors are cash-on-delivery, fraud exposure and order value.

• Justified when payment is on delivery, where an unconfirmed order means a delivery attempt, a return journey and unsold stock. In markets where cash-on-delivery is common this is the single strongest case for outbound calling in ecommerce.
• Justified on high-value orders where an error or a fraudulent order is expensive to unwind.
• Justified where the order contains something that needs clarification — a size, a variant, a delivery constraint, an address that does not resolve.
• Not justified for routine prepaid orders, where the confirmation email does the job and a call reads as suspicion.
• The call should confirm the items, the address and the delivery window, answer questions, and offer to change or cancel — a cancellation now is cheaper than a failed delivery later.
• Every outcome writes back to the order, including 'unreachable', which should have its own defined handling rather than proceeding by default.

Use case 2: abandoned checkout

The most commonly proposed and most easily overdone. Calling someone who left a basket is intrusive unless the circumstances justify it.

1. 1

   Apply a value threshold, and set it high

   Most abandoned baskets are not worth a call and calling about them trains customers to associate your brand with unsolicited contact. Only the top band by value should be candidates.

2. 2

   Try messaging first, always

   An email and a message cost almost nothing and recover a meaningful share. The call is for what remains after those have failed, not the first response.

3. 3

   Only call known customers

   Someone with a purchase history and a phone number they have given you in a prior transaction is a different proposition from a first-time visitor who entered a number into a checkout field.

4. 4

   Frame it as help, not as sales

   'I noticed the order did not complete — was there a problem with payment or delivery?' This is genuinely useful, because a real share of abandonments are payment failures or delivery-option problems the customer could not resolve.

5. 5

   Capture the reason, whatever the outcome

   Why checkouts fail is information most retailers do not have, and it frequently identifies a fixable problem in the checkout itself — which is worth more than the recovered orders.

6. 6

   Stop at one attempt

   A second call about an abandoned basket is pursuit. One call, then messaging, then stop.

The reason-capture point is the strongest argument for this use case. A dozen calls revealing that a payment method fails at a particular step is worth more than the baskets recovered.

Use case 3: delivery problems

The least contentious and most appreciated of the three. Customers rarely object to a call about a problem with their order.

• Delivery delayed or failed: call proactively, before the customer discovers it. This converts a likely complaint into a demonstration of competence and is consistently under-used.
• Address unresolvable or incomplete: a call resolves in a minute what a message exchange takes a day to resolve.
• Item out of stock after ordering: offer alternatives, a wait, or a refund, and let the customer choose rather than deciding for them.
• Delivery attempted and missed: arrange the next attempt in the call rather than leaving a card.
• Return or exchange in progress with a problem: worth a call because the customer is already dissatisfied.
• Not for routine tracking updates, which messaging handles better and which customers prefer not to be called about.

Proactive problem calls are the use case with the clearest customer benefit and the lowest intrusion risk. If you are only going to build one of the three, this is usually the one.

Call or message: the decision

Applying this consistently prevents the programme from expanding into contact customers do not want.

• Message when the purpose is to inform: dispatch, tracking, delivery confirmation, promotional content. Always.
• Message first for anything recoverable, and call only when it fails.
• Call when a decision is needed from the customer and waiting is costly — a substitution, a delivery window, a cancellation.
• Call when there is a problem you caused.
• Call when money is at risk in a way messaging will not resolve in time.
• Never call about a promotion or an offer without a prior relationship and a clear basis for doing so.
• Honour channel preference where the customer has expressed one, and record it the first time.
• Never call and message simultaneously about the same thing.

Data and integration requirements

These calls depend on live commerce data more than on conversation design.

• Real-time order state: items, payment status, fulfilment stage, delivery attempts. A call referencing stale order data is worse than no call.
• Stock availability, so alternatives offered are genuinely available.
• Customer history, to distinguish known customers from first-time visitors.
• Checkout abandonment events with value and contents, ideally as real-time events rather than a batch export — the triggered call pattern.
• Delivery and carrier status where problems originate outside your systems.
• The ability to act during the call: cancel, amend, substitute, rebook delivery, or at minimum record the request reliably.
• Write-back to the order and to the customer record.
• Calling-time rules per market, and suppression including opt-outs and frequency caps across all outbound activity — see retry strategy.

The ability to act during the call is what separates a useful call from a message read aloud. Our web platforms page covers building this integration, there is an online store project in our portfolio, and a store integration removes part of the plumbing where you run on a standard platform.

Metrics

Each use case has a different measure and they should not be blended.

1. 1

   Order confirmation: failed delivery rate

   The point of confirming is to avoid a delivery that cannot be completed. Measure that, not calls made, and compare against a pre-launch baseline.

2. 2

   Abandoned checkout: incremental recovery

   Against a control group that received only messaging. Without a control, the calls will be credited with recoveries that messaging would have produced anyway.

3. 3

   Delivery problems: complaint rate and resolution time

   Proactive calling should reduce inbound complaints about the same issues. That reduction is the measure.

4. 4

   Across all three: opt-out and complaint rate

   The counterweight. If outbound contact is rising and so are opt-outs, the programme is borrowing from future goodwill.

The control group for abandoned checkout is the discipline that keeps this honest. It is easy to run and it frequently shows a smaller effect than the uncontrolled figure suggests.

Failure modes

Ecommerce outbound fails in recognisable ways.

• Calling about routine matters a notification handles.
• Calling every abandoned basket regardless of value.
• Calling first-time visitors who entered a number into a checkout field.
• Referencing stale order data.
• Offering alternatives that are out of stock.
• Calling and emailing about the same thing at once.
• No ability to act during the call, so it ends in 'someone will sort this out'.
• Repeated attempts about an abandoned basket.
• Ignoring stated channel preferences.
• Measuring abandoned-checkout recovery without a control group.

Where calling does not belong

Some ecommerce outbound should not happen at all.

• Promotional calling without a clear prior basis.
• Upselling during a problem call, which converts goodwill into resentment.
• Calling about returns in a way that discourages them.
• Anything involving a payment dispute, which needs a person.
• Calls to customers who have complained recently.
• Any market where calling rules for your situation have not been confirmed by whoever owns that decision in your business.
• Age-restricted or sensitive product categories, where an unsolicited call may be inappropriate regardless of the commercial logic.

The second one is worth stating plainly: a call about a delivery failure that ends with an offer is remembered as a sales call. Solve the problem and stop.

Decision framework and next step

Four questions.

1. 1

   Is cash-on-delivery a meaningful share of your orders?

   If so, confirmation calling has clear arithmetic behind it. If not, it probably does not.

2. 2

   What does messaging already recover?

   Establish this before adding calls, or the calls will be credited with it.

3. 3

   Can you act during the call?

   Cancel, amend, substitute, rebook. Without this, a message is better.

4. 4

   What is your current inbound complaint volume about delivery problems?

   That number is the case for proactive calling, and it is usually larger than expected.

Start with proactive delivery-problem calls — clearest benefit, lowest intrusion, and the easiest to justify internally. Add confirmation calling if cash-on-delivery makes the arithmetic work, and treat abandoned checkout as a narrow, high-value, message-first exception. The on-site equivalent is covered in product recommendation chatbots; our AI solutions overview covers staging.

Frequently asked questions

1. 1

   When is an outbound call justified in ecommerce?

   When there is a decision to be made, a problem to solve, or money at risk, and a message has failed or would be too slow. Ecommerce is a messaging business, so calling works precisely because it is rare.

2. 2

   Should abandoned checkouts be called?

   Only above a high value threshold, only for known customers with a prior relationship, only after messaging has failed, and only once. Framed as help rather than sales — a real share of abandonments are payment or delivery problems the customer could not resolve.

3. 3

   What is the strongest use case?

   Proactive calls about delivery problems: delays, failed attempts, stock issues, unresolvable addresses. Customers rarely object, it converts likely complaints into demonstrations of competence, and it is consistently under-used.

4. 4

   What integrations are needed?

   Real-time order state, stock availability, customer history, abandonment events, carrier status, the ability to cancel or amend during the call, write-back to the order, and suppression with calling-time rules.

5. 5

   How should it be measured?

   Per use case: failed delivery rate for confirmation calls, incremental recovery against a messaging-only control for abandoned checkout, complaint rate and resolution time for delivery problems — with opt-out rate as the counterweight across all three.

The narrow version works: call about problems, confirm where the economics require it, and leave everything else to messaging.